The degradation of electromechanical systems is a continuous process that must be prevented. In order to minimize losses and the increase of maintenance costs, conditioning monitoring should be applied to critical equipment (Condition Based Maintenance-CBM)[1]. Conditioning monitoring can use statistical methods to estimate the present state of the collected data. The Accelerated Failure-Time Models can be an important method for prediction system failures. In this paper collected data are vibration values, obtained with overall vibration values (RMS) from four points from an electropump, and in a second phase, a failure will be induced to degrade the equipment state. When variables are not independent, it is suggested modeling data with ARIMA (p, d, q) models [2], and use the residues for the first Phase. In the Phase I, the estimation of parameters is achieved using the traditional Shewhart charts. The combination of the four variables, T2, is performed only after the ARIMA model is estimated. The special multivariate control charts are used to predict failures. Assuming independence and Normal Distribution, MEWMA Modified Control Charts [3] will be applied in the Phase II to new collected data, by this procedure, behavior of the equipment will be controlled. In phase II, if the data are dependent, the prediction errors are used. The previous methodology is used just for fault detection. Using a chart expressing vibration as a function of velocity (m/s) it will probably possible to make diagnosis and identify which component is causing vibration. Using these methodologies allows online condition monitoring and act in time to minimize maintenance and outage costs.